Exposure variables

The definition for overweight is having a BMI greater than or equal to 25 kg/m² and below 30 kg/m². The definition for obesity is having a BMI greater than or equal to 30 kg/m² [25, 26]. According to the World Health Organization (WHO), the definition for abdominally overweight or obesity is a WC of greater than or equal to 80 cm and 88 cm, respectively, for females, and 94 cm and 102 cm, respectively, for males [25, 26].

Disease outcomes

Possible co-morbidities of overweight and obesity were identified from a preliminary search reviewed by an eating disorder and obesity expert and a review of previous systematic reviews [1–3]. We also reviewed the leading causes of global burden of disease and included the diseases reported with burden attributable to overweight and obesity [27]. Twenty co-morbidities were initially included in this analysis: cancer (kidney, colorectal, prostate, ovarian, uterine/endometrial, esophageal, pancreatic, and post-menopausal breast), type II diabetes, cardiovascular disease risk (hypertension, coronary artery disease, congestive heart failure, pulmonary embolism, stroke, dislipidaemia), gallbladder disease, chronic back pain, osteoarthritis, asthma, and sleep apnea.

Literature review

A literature search was conducted using the search terms: 'Incidence, Prevalence, Risk, Risk Factors, Cohort Studies, Longitudinal Studies, Follow-up Studies, or Prospective Studies' in combination with 'Adipose Tissue, Obesity, Body Mass Index, or Body Composition' (all "exploded"). These same search terms were applied to each co-morbidity (also "exploded") for both Medline and Embase search engines to retrieve all potentially relevant English articles (until January, 2007). We also searched ISI Web of Science, Google Scholar, and the bibliographies of retrieved articles.

The articles obtained from the literature search were then evaluated according to criteria set out in Figure 1. Criteria for inclusion were: prospective cohort study of the general population of a Western country (countries in Europe or North America, Australia or New Zealand), relevant outcomes, a sample size of at least 200 subjects, and risk estimate based on the incidence of disease instead of the mortality rate of disease. For large cohorts with multiple articles meeting the defined criteria, the most recent article or the article with the most usable information was used. Studies were excluded if they did not provide enough data to allow calculation of unadjusted relative risks (RRs) with 95% confidence intervals (CI) for the overweight and obese groups compared to the normal group. Data extracted for study characteristics included study design, country, cohort name, duration of follow-up, number of patients in each study group, age range, gender and ethnicity. Ascertainment of exposure and outcome variables was also recorded. The literature search was conducted by ZA, the decisions on inclusion and exclusion were made by ZA, NB, DPG, CLB and the data were extracted by ZA and DPG.

Meta-analysis

RRs were measured by incidence rate ratios (IRRs) when person-time data were available and by the ratios of proportions (RR-Ps) when person-time data were not available. Study-specific unadjusted RRs on the log scale comparing overweight with normal and obese with normal were weighted by the inverse of their corresponding variances to obtain a pooled RR with 95% CIs. We used the random-effects model to estimate the pooled RR using the maximum likelihood estimation method [28]. The Q statistic was also calculated to assess the homogeneity of RRs (log scale) [28]. Potential publication bias was visually inspected by funnel plots and tested by asymmetry tests [29, 30]; however it was performed only on those meta-analyses which included sufficient number of studies (N>5). We also did various sensitivity analyses stratified on the length of follow-up, age criteria and country to examine the robustness of the results.

WC measurements were considered to be the better risk predictor for some co-morbidities such as diabetes, hypertension, coronary artery disease, congestive heart failure, stroke and gallbladder disease [16–24]. Therefore, when the RRs based on WC measurements for these co-morbidities were available, they were pooled separately from those based on BMI and used as the final RRs for these co-morbidities. When both IRR and RR-P estimates were available, both estimates were presented. Final results of RRs were selected based on the number of pooled studies, the duration of study follow-up and the sample size of included studies

Breast cancer

In total, 14 studies were identified for post menopausal breast cancer (Figure 2) [19, 32–44]. The pooled IRRs [95% CI] across categories of WC were 1.13 [1.01–1.07] for overweight and 1.30 [1.17–1.44] for obesity while across categories of BMI the IRRs were 1.08 [1.03–1.14] for overweight and 1.13 [1.05–1.22] for obesity.

Endometrial cancer

In total, 10 studies were identified to meet the inclusion criteria for endometrial cancer (Figure 3) [19, 31, 35, 36, 42, 45–49]. The study on the US black women was not included in the final result as it showed systematic difference from other studies mainly on Caucasian population [31]. The pooled IRRs [95% CI] across categories of WC were 1.15 [1.02–1.30] for overweight and 1.42 [0.80–2.49] for obesity while across categories of BMI IRRs were 1.53 [1.45–1.61] for overweight and 3.22 [2.91–3.56] for obesity.

Ovarian cancer

In total, 9 studies were identified for ovarian cancer (Figure 4) [19, 35, 36, 42, 45, 50–53]. The single IRR [95% CI] across categories of WC were 0.61 [0.35–1.08] for overweight and 1.35 [0.95–1.93] for obesity while pooled IRRs estimates across categories of BMI were 1.18 [1.12–1.23] for overweight and 1.28 [1.20–1.36] for obesity.

Colorectal cancer

A total of 12 studies were identified for colorectal cancer (Figures 5 and 6) [19, 42, 45, 54–62]. For men, the pooled IRRs [95% CI] across categories of WC were 1.88 [1.47–2.41] for overweight and 2.93 [2.31–3.73] for obesity and across categories of BMI were 1.51 [1.37–1.67] for overweight and 1.95 [1.59–2.39] for obesity. For women, the pooled IRRs across categories of WC were 1.25 [0.98–1.59] for overweight and 1.55 [1.27–1.88] for obesity and those across categories of BMI were 1.45 [1.30–1.62] for overweight and 1.66 [1.52–1.81] for obesity.

Esophageal cancer

Only 1 study was identified for esophageal cancer [63]. The study found the risk of cancer based on overweight to be 1.15 [0.97–1.36] and 1.13 [1.02–1.26] for females and males, respectively. The corresponding risks for obesity were 1.20 [0.95–1.53] and 1.21 [0.97–1.52].

Kidney cancer

We identified 5 studies meeting the inclusion criteria relating overweight and obesity to kidney cancer (Figure 7) [42, 45, 57, 64, 65]. The pooled IRRs [95% CI] across categories of BMI for men were 1.40 [1.31–1.49] for overweight and 1.82 [1.61–2.05] for obesity. For women the corresponding risks were 1.82 [1.68–1.98] and 2.64 [2.39–2.90].

Pancreatic cancer

The search identified 6 studies giving information on the risk of pancreatic cancer attributable to overweight and obesity (Figure 8) [42, 45, 66–69]. The pooled IRRs [95% CI] across categories of BMI for men were 1.28 [0.94–1.75] for overweight and 2.29 [1.65–3.19] for obesity. For women the corresponding risks were 1.24 [0.98–1.56] and 1.60 [1.17–2.20].

Prostate cancer

The search identified 8 studies giving information on the risk of prostate cancer attributable to overweight and obesity (Figure 9) [36, 42, 45, 70–74]. The pooled IRRs [95% CI] across categories of BMI were 1.14 [1.00–1.31] for overweight and 1.05 [0.85–1.30] for obesity.

Type II diabetes

Nine studies met the inclusion criteria and were included in the meta-analysis (Figure 10) [19, 75–82]. In general, elevated BMI and WC were significantly associated with type II diabetes in men and women. The pooled IRRs [95% CI] across categories of BMI were 2.40 [2.12–2.72] and 6.74 [5.55–8.19] in men while the corresponding IRRs in women were 3.92 [3.10–4.97] and 12.41 [9.03–17.06]. The association between increased WC and type II diabetes was similar but weaker in comparison with BMI. Only two studies were included in men. The pooled IRRs [95% CI] across categories of WC were 2.36 [1.76–3.15] and 5.67 [4.46–7.20] in men and the pooled RR-Ps [95% CI] based on the same two studies were 2.27 [1.67–3.10] and 5.13 [3.81–6.90], respectively. The pooled RR-Ps were more conservative RR estimates and presented in the summary table (Table 2). The pooled IRRs [95% CI] across categories of WC were 3.40 [2.42–4.78] and 11.10 [8.23–14.96] in women.

Hypertension

Four studies met the inclusion criteria and were included in the meta-analysis (Figure 11) [19, 83–85]. The pooled IRR [95% CI] estimates for hypertension across categories BMI for men were 1.28 [1.10–1.50] for overweight and 1.84 [1.51–2.24] for obesity. The corresponding figures for females were 1.65 [1.24–2.19] and 2.42 [1.59–3.67]. The single IRR estimate based on WC for women was 1.38 [1.27–1.51] for overweight and 1.90 [1.77–2.03] for obesity.

Stroke

Seven studies met the inclusion criteria and were included in the meta-analysis (Figure 12) [86–92]. The pooled RR-P [95% CI] estimates for stroke across categories BMI for men were 1.23 [1.13–1.34] for overweight and 1.51 [1.33–1.72] for obesity. The corresponding results for females were 1.15 [1.00–1.32] and 1.49 [1.27–1.74].

Coronary Artery Disease

Eleven studies were identified with evidence for coronary artery disease related to obesity (Figures 13 and 14) [20, 21, 79, 91, 93–99]. The pooled RR-P estimates for coronary artery disease across categories of WC were more conservative RR estimates for men than the corresponding IRR estimates and thus were presented in the summary table (Table 2). The RR-P [95% CI] estimates for WC were 1.41 [1.16–1.72] for overweight and 1.81 [1.45–2.25] for obesity. The corresponding results for BMI were 1.29 [1.18–1.41] and 1.72 [1.51–1.96]. While the pooled RR-P estimates based on BMI measurements for women were generated from 4 studies, the IRR estimates were generated from 2 different studies with longer follow-up. Thus, the IRR estimates were the RR estimates for women presented in the summary table (Table 2). The estimates were 1.80 [1.64–1.98] for overweight and 3.10 [2.81–3.43] for obesity based on BMI measurements and 1.82 [1.41–2.36] and 2.69 [2.05–3.53] for WC.

Congestive Heart Failure

Four studies were identified with evidence for congestive heart failure related to obesity (Figure 15) [91, 100–102]. The pooled IRR [95% CI] estimates for congestive heart failure across categories of BMI for men were 1.31 [0.96–1.79] for overweight and 1.79 [1.24–2.59] for obesity. The pooled RR-P estimates for females based on 3 studies were 1.27 [0.68–2.37] and 1.78 [1.07–2.95], which were chosen as the RR estimates over the IRR estimates based on 2 studies.

Asthma

Four studies were identified with evidence for asthma related to obesity (Figure 16) [103–106]. The pooled RR-P [95% CI] estimates for asthma across categories of BMI for men were 1.20 [1.08–1.33] for overweight and 1.43 [1.14–1.79] for obese. The corresponding numbers for females were 1.25 [1.05–1.49] and 1.78 [1.36–2.32].

Chronic back pain

Only 1 study was identified to meet the inclusion criteria for chronic back pain [107]. The study identified the association for the overweight and obesity with early retirement due to chronic back pain. The study found RR-P [95% CI] estimates across categories of BMI of chronic back pain based on overweight to be 1.59 [1.34–1.89] and for obesity 2.81 [2.27–3.48].

Osteoarthritis

We identified three studies meeting the inclusion criteria relating overweight and obesity to osteoarthritis (Figure 17) [108–110]. The studies identified the risk of joint replacement attributable to being overweight and obese. The pooled IRRs [95% CI] across categories of BMI for men were 2.76 [2.05–3.70] for overweight and 4.20 [2.76–6.41] for obesity. For women, the RR-P estimates were more conservative RR estimate and they were 1.80 [1.75–1.85] and 1.96 [1.88–2.04].

Pulmonary embolism

Only 1 study was identified to meet the inclusion criteria for pulmonary embolism [111]. The study found IRR [95% CI] across categories of BMI of Pulmonary embolism based on overweight to be 1.91 [1.39–2.64] and for obesity 3.51 [2.61–4.73].

Gallbladder disease

We identified four studies meeting the inclusion criteria relating overweight and obesity gallbladder disease (Figure 18) [59, 112–114]. The pooled IRRs [95% CI] across categories of WC for men were 1.63 [1.42–1.88] for overweight and 2.51 [2.16–2.91] for obesity. The corresponding pooled RR-Ps were 1.61 [1.40–1.85] and 2.38 [2.06–2.75]. Thus, the pooled RR-Ps were more conservative RR estimates and presented in Table 2. Across categories of BMI, the pooled IRR and RR-P estimates were presented as the RR estimates in Table 2 and they were 1.09 [0.87–1.37] for overweight and 1.43 [1.04–1.96] for obesity. For women only estimates for BMI were identified giving pooled IRR and RR-P estimates of 1.44 [1.05–1.98] for overweight and 2.32 [1.17–4.57] for obesity.

Potential publication bias was assessed for post-menopausal breast cancer, endometrial cancer, ovarian cancer, colorectal cancer, pancreatic cancer for females and prostate cancer. We found some evidence of funnel-plot asymmetry for obesity in prostate cancer where bigger studies tended to show stronger positive association than smaller studies. No evidence of publication bias was found in the other meta-analyses.

Our sensitivity analyses showed that our results were in general robust with the following exceptions. For ovarian cancer, associations for both overweight and obesity were slightly weaker in US studies compared to European studies. Similar country differences were found in pancreatic cancers; in addition, weaker associations were observed in older population. Studies with shorter follow-up time and of older population showed slightly weaker association of obesity with prostate cancer. In coronary artery disease for females, studies with shorter follow-up showed weaker associations of both overweight and obese. In coronary artery disease for males, weaker associations were observed in US and Canadian studies. Studies of post-menopausal and senior women on congestive heart failure showed weaker associations for both overweight and obesity.